docs: clarify agent output and trajectory concepts

[minixalpha]

Summary

• Rework the bilingual agent-output research notes so run output, execution trace, benchmark trajectory, and resumable session are defined before the product comparison.
• Clarify single result objects, public JSONL/NDJSON event streams, partial and delta content, terminal events, and stdout transport formats, with concrete examples and separate comparison tables.
• Compare Pi, Claude Code, Codex, OpenCode, and Grok Build across public output, diagnostics, evaluation artifacts, and resumable state, then refine the proposed nanoPyCodeAgent CLI contracts.
• Add a Code Agent Runtime glossary in CONTEXT.md to preserve the terminology boundaries.
• Regenerate the complete English research note from the updated Chinese source of truth.

Validation

• git diff --check
• Verified that the Chinese and English notes have matching heading, code-fence, table-row, and link counts, with identical external URL sets.
• Documentation-only change; no automated code tests were run.

Bilingual research status

• The Chinese research note is the source of truth.
• The corresponding English research note has been fully regenerated and is in sync.